KDD 2026 Recap: Jeju, Keynotes, Tracks, 2027 Prep

KDD 2026, the ACM SIGKDD Conference on Knowledge Discovery and Data Mining, wrapped August 9-13, 2026 at the International Convention Center Jeju on Jeju Island, South Korea. Workshops and tutorials ran August 9-10; the main conference and all three keynotes ran August 11-13. Jeff Dean (Google Chief Scientist), Jingren Zhou (Alibaba Chief AI Architect), and Regina Barzilay (MIT) held the keynote stage. Five submission tracks, no vendor floor pretending to be an agenda (checked 2026-08-24, kdd2026.kdd.org).

I track KDD because the techniques debated on academic stages here tend to ship in production AI systems 12 to 24 months later. Most events I cover are commercial summits with a sales pitch buried in every keynote. KDD is the SIGKDD flagship, running continuously since 1995, and a large chunk of the data mining and machine learning methods your team already uses first landed as peer-reviewed KDD papers.

Why KDD 2026 mattered more than the keynote roster suggests

Google Chief Scientist. Alibaba Chief AI Architect. MIT AI-for-medicine professor. Three keynotes covering three separate applied surfaces. That is the marketing headline. The load-bearing signal was underneath it.

KDD 2026 accepted work into five review-hardened tracks: Research, Applied Data Science, Datasets and Benchmarks, AI for Sciences, and Blue Sky Ideas. Each track ran its own review committee. That structure is the actual product. A commercial AI summit hands you 300 talks curated by a marketing team; KDD 2026 handed you an accepted-paper list built by peer reviewers who send papers back for revision. The Datasets and Benchmarks Track alone matters more to production AI teams than any keynote. If a benchmark passes KDD peer review, you can trust the leaderboard your model is measured against.

The keynote that actually moved

Jeff Dean opened Tuesday, August 11 with the trends keynote. Jingren Zhou followed Wednesday on the agentic data stack: how large language models are reshaping data engineering and orchestration inside Alibaba’s infrastructure. Regina Barzilay closed Thursday on rethinking disease diagnosis and treatment.

Zhou’s talk is the one operators cared about. Agentic orchestration inside a hyperscaler’s data engineering pipeline is exactly the pattern SaaS teams are trying to figure out without the hyperscaler’s engineering bench. If you missed the room, the workshop and tutorial recordings are where the practical follow-up lives, not the keynote video.

KDD Cup 2026 ran alongside the main conference

The KDD Cup has run since 1997, longer than any comparable data-mining competition. The 2026 edition opened around March 1, notified winners July 15, and hosted the winner presentations at a dedicated workshop on Wednesday, August 12. Recent Cups have carried prize pools around $30,000 per organizing track (checked 2026-08-24, kdd2026.kdd.org).

Practitioners who never submit a paper still show up for the Cup because the winning solutions are peer-scrutinised and ship as GitHub repos within weeks. Watch the KDD 2026 Papers page for the write-ups, not the aggregator recaps.

KDD 2026 pricing, for the record

Registration is closed. The pricing tiers matter for anyone planning KDD 2027 travel and budget approval now.

Tier	Non-Member	SIGKDD Member	Student
Early Bird (through June 17)	$1,250	$1,000	$500
Standard (through August 8)	$1,500	$1,250	$600
Onsite (from August 9)	$1,750	$1,500	$700

(All USD, checked 2026-08-24, kdd2026.kdd.org registration page.)

The gap between early-bird and onsite is $500 flat per tier. That is real money, not a marketing pressure tactic. A SIGKDD or partner-society (SIGMOD, AAAI, general ACM) membership knocks another $250 off every tier and often costs less than the discount itself. Student pricing is legitimately fair for a five-day peer-reviewed conference. If you are a graduate student presenting, the Student Travel Award covers a meaningful chunk.

Where to find the actual accepted papers

The accepted paper list is the honest signal for whether a track justified the flight, and it is what promotional pages bury. KDD 2026 published its accepted papers on the official Papers page at kdd2026.kdd.org/papers/, with peer review run through OpenReview. Read the titles in your specific subfield before you read a single keynote bio. Secondhand roundups lag the official source by weeks and get the tracks wrong.

My verdict, for KDD 2027 planning

For working data scientists, ML researchers, and applied industry teams shipping knowledge discovery work, KDD stays the most legitimate research conference in the calendar. Peer review is the moat. A keynote lineup with Google’s Chief Scientist, Alibaba’s Chief AI Architect, and an MIT professor is not built with a marketing budget. It is built on 30-plus years of research credibility.

You should send a person to KDD 2027 if you or your team have a paper, a case study, or a genuine research question that fits the Research Track, the Applied Data Science Track, or a Special Day. You should send one person to scout if budget is tight and you want a read on where applied AI research is heading before committing more. You should skip it if your interest in AI is general rather than research-grade. There are plenty of commercial AI events built for that audience, and KDD is not trying to be one of them.

The Applied Data Science Track and the Special Days (Finance, Health, Education) are where industry practitioners get the most direct return, not the keynote stage alone. Anyone can watch a keynote recording later. The applied sessions built around your specific domain are the part worth flying for.

KDD 2027 dates, venue, and call for papers are not published yet. When they land, they will show up on kdd.org first, not on the aggregator sites. Watch KDD Cup 2027 announcements and the KDD 2027 call for the honest calendar, and see how KDD’s pipeline compares against SIGKDD 2027 programming when both are locked.